The Model Context Protocol (MCP) provides a standardized way for AI applications to discover and execute external tools. While local MCP servers communicate via subprocesses, shifting to remote, internet-exposed servers introduces significant security risks—specifically, the potential for unauthorized tool execution, requiring robust identity and permission management to prevent data exposure.

Adopting MCP streamlines tool integration, but exposing these tools over the internet mandates moving beyond simple API keys toward identity-aware, scoped authorization. Unless your tools are strictly personal and local, treating the remote endpoint as a public interface and implementing OAuth-based user isolation is a requirement for production safety.